AI breakthrough optimises electric fleet routing in complex urban environments

Researchers have developed a new deep reinforcement learning method to tackle the Electric Dial-a-Ride Problem (E-DARP). This approach optimises routing, charging, and service quality for electric vehicle fleets in complex urban air environments. Early results show it outperforms traditional techniques in both speed and efficiency.

The team's solution combines a graph neural network encoder with an attention-driven route construction policy. Unlike older methods, it directly processes real-world factors like travel time, energy use, and asymmetric road costs. These features allow it to handle non-Euclidean routing challenges in dynamic urban settings.

Testing on standard benchmarks revealed the method matched the best-known results within 0.4% while cutting computation times. It also proved effective on large-scale scenarios with up to 250 request pairs, accounting for nonlinear charging and realistic energy constraints. Robustness experiments confirmed its reliability under unpredictable conditions.

Further sensitivity analyses measured how battery capacity, fleet size, and reward weights affected performance. The findings highlight the policy's adaptability across different operational demands. By jointly optimising routing, charging, and service quality, the approach delivers a scalable solution for modern electric fleets.

This breakthrough offers a practical way to manage electric vehicle fleets in cities, balancing efficiency with real-world constraints. The method's speed, accuracy, and flexibility could help operators reduce costs while maintaining service standards. Further testing may expand its applications in urban logistics.